	US 11,995,523 B2
	Systems and methods for determining training parameters for dialog generation
Omar Florez Choque, Oakland, CA (US); Erik T. Mueller, Chevy Chase, MD (US); and Zachary Kulis, Arlington, VA (US)
Assigned to CAPITAL ONE SERVICES, LLC, McLean, VA (US)
Filed by Capital One Services, LLC, McLean, VA (US)
Filed on Jan. 27, 2021, as Appl. No. 17/159,255.
Application 17/159,255 is a continuation of application No. 16/669,866, filed on Oct. 31, 2019, granted, now 10,929,781.
Prior Publication US 2021/0150414 A1, May 20, 2021
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 20/00 (2019.01); G06F 3/16 (2006.01)

CPC G06N 20/00 (2019.01) [G06F 3/16 (2013.01)]

20 Claims

1. A method for training artificial dialog system to have human-like customer service interactions, the method comprising:

receiving, at a processor, a first input from a first user;

receiving, at the processor, a first response to the first input from a second user;

determining, by the processor executing a training algorithm, a first intent for the first input based at least in part on the first input and the first response;

identifying, by the processor executing the training algorithm, a first action based on the first input, the first response, the first intent, or some combination thereof;

determining, by the processor executing the training algorithm, one or more first trainable parameters based on the first input, the first response, the first action, the first intent, or some combination thereof;

determining, by the processor, whether the one or more first trainable parameters are positive or negative;

updating, by the processor, the training algorithm based on the one or more first trainable parameters;

wherein determining whether the one or more first trainable parameters are positive or negative comprises:

receiving a survey from the second user;

assigning a value to each of one or more entries in the survey;

generating a composite score based on a sum of the values of each of one or more entries; and either:

determining the one or more first trainable parameters are positive when the composite score of the survey is above a numerical value; or

determining the one or more first trainable parameters are negative when the composite score of the survey is below the numerical value.